Roadmap

Telephony — inbound and outbound calling

Today, a Wetel agent is reachable through the embeddable web SDK (<vai-avatar>) or through a custom UI you build against the GraphQL API. Live voice calls from a browser or an app (sdkStartVoiceCall) work on the staging environment, on request; they are not in production. There’s no phone-network integration: an agent can’t currently receive or place a real phone call, on a Malaysian (+60) number or any other, and it can’t join an online meeting. See What works today, what does not, what is next.

Calls are planned inbound first, through a licensed telecom provider. Outbound calling is limited by policy to callbacks the person asked for and follow-ups with people who agreed to be called (specification stage); Wetel does not offer cold outbound sales calls.

This roadmap item extends the same agent and workflow you already configure to a phone-call channel. The persona, the workflow graph, the session/message model — none of that changes. What changes is the channel a session comes in on: instead of (or alongside) a web embed, a call would start a session the same way, run the same workflow, and produce the same message history you already query today. If you’ve already built a workflow for chat or voice-over-web, the goal is that it works for a phone call with no redesign.

Voice: long recordings, streaming and custom vocabulary

Speech-to-text and text-to-speech are available today, including mixed-language (Malaysian “rojak”) speech: see the speech guide. Voice notes and short utterances are the supported use. Clips up to about five minutes are handled, one request per clip.

What is still ahead, none of it available today:

Long recordings with speaker labels, for interviews and call recordings, so a conversation comes back as who said what.
Streaming recognition, for live calls where text must arrive while the person is still speaking. This is also the first step towards the telephony item above.
Custom vocabulary, a list of names, products and local words you want recognised correctly.
Accuracy and fallback monitoring, so you can see how often each speech engine answered and when the platform fell back to the other one.

Hiring a role yourself

Wetel already ships role templates: ready-made AI roles (a persona, a conversation flow and the knowledge bases and connections it needs) that Wetel sets up for you in your workspace, with a setup checklist for anything that still needs your own credential. See agentTemplateInstall and the changelog.

Workspace owners can also hire a role themselves (owner only, PRO plan for now, with limits on roles at once and hires per day): agentTemplates lists the roles Wetel has made available and hireAgentTemplate hires one. Nothing appears in that list until Wetel lists a role, so today it is available by invitation. What is still ahead is a way to move an existing hire to a newer version of its template, and a wider catalogue. The create, update and publish flow underneath does not change.

Event-based triggers

Scheduled workflows are available today: a workflow can start on a schedule, with no message involved. See Scheduled workflows.

What is still ahead is the event side: starting a workflow because something happened (“when a record changes, run this graph”) rather than because a clock fired. The graph, nodes and execution trace stay exactly as they are today; only what starts a run changes.

Supervision: trends and alerts

The per-agent view of what an agent handled is available today: agentActivitySummary rolls up runs by status, recovered failures, average evaluation score, open questions and schedule health for each agent, and the dashboard shows it as one card per agent.

What is still ahead is the view over time and the alerting around it: how those numbers trend, credits per agent, and being told when an agent needs a person without having to look.

What’s already shipped

This page is intentionally forward-looking. For what’s already built and live, see the Changelog — it tracks real, shipped API changes as they land.